Good work!
I've just got round to using the new version but it seems that mode 14800 iTunes back up>=10 is a lot slower then previously. I don't have time right now to get more specifics but so far, I seem to only get around 30% compared to using 3.6
The solution is to use -w 4 (which enables all threads on the GPU) but has other disadvantages. So use it for -m 14800 only and for the other modes use -w 3.
Excellent, looks like thats worked out. Was there a reason for this slight change?
Notice the change in the maximum password length.
True @ password length, but for this change the reason was fine tuning. In general it's better to have a low number of threads, as long as it doesn't drop the performance. For NV the ideal OpenCL thread number is the warp size (and for AMD the wavefront number). The lower the thread count, the smaller the workload, the faster the restore checkpoint and other things. There's a few GPU which in combination with some particular hash modes do not scale that good on a low thread count and 14800 is one of them.
Ahh ok, very good. I knew about the max password length but wasn't sure how that would affect the speed so much. Good to know